floatboat
Open navigation menu

Orca

Orca brings multiple coding agents into one workspace, with separate Git worktrees and a shared review interface. Study task isolation, parallel comparisons, and diff feedback to organize agent work without confusing simultaneous execution with a verified result.

Plan parallel changes without mixing their worktrees

Describe repository work that could benefit from separate attempts or independent review. The request can compare Orca-style ownership and worktree boundaries, with a fixed handoff revision, shared checks, and a distinct decision about which results should be integrated.

Describe frontend, backend, and test boundaries, shared files, and merge dependencies.

Copy the request into a new Floatboat conversation and provide the required material.

Download Floatboat

Your request for Floatboat

Preview and copy the request. It is not sent automatically from this page.

Reference: Orca
https://github.com/stablyai/orca

Read the original documentation and check the current environment and dependencies. Propose any new installation or tool connection before proceeding.

My task:
Study Orca’s worktree documentation to plan parallel frontend, API, and test work for my search feature. Identify shared-file conflicts, branch ownership, merge order, and end-to-end checks. Show the plan before creating worktrees or launching other agents.

Expected deliverable:
A worktree/branch plan with task ownership, integration order, and a combined verification checklist.

About Orca

Orca is a development workspace published in stablyai/orca. It organizes existing coding agents such as Codex, Claude Code, OpenCode, and Pi alongside terminals, repository files, and review tools. It coordinates agent processes rather than supplying a new foundation model.

Each parallel task can use its own Git worktree. That separates file changes while the workspace provides a place to inspect diffs and send feedback. A mobile companion and SSH workflows extend how work is monitored or executed, but model access and remote-environment setup remain separate requirements.

For a Floatboat workflow, the useful pattern is to define bounded assignments, isolate changes, and compare evidence before accepting an implementation. Parallel agents still need a shared task contract and an explicit integration owner. Running in separate worktrees does not itself prevent mistakes, conflicts, or unauthorized actions.

Coordinate Independent Attempts and Review Their Results

Orca separates file isolation, review before integration, and observation across environments. Compare those responsibilities when coordinating agents, with clear task ownership and shared checks so parallel execution does not obscure who decides which changes enter the final result.

  • Isolate file changes by task

    Assign worktrees to bounded tasks or competing approaches, so one agent does not silently overwrite another agent’s uncommitted files.

  • Review changes before integration

    Use diffs and line-level comments to turn review findings into focused follow-ups, then inspect the revised result.

  • Observe work across environments

    Desktop, companion, and SSH interfaces support different monitoring or execution arrangements. Choose the one that fits the actual environment.

Use Cases for Orca

Competing implementations, independent review, and parallel repository edits need comparable baselines. Explore Orca's coordination patterns through fixed revisions, isolated worktrees, and shared acceptance checks, with ownership and final integration decisions defined before work begins.

Compare two implementation approaches

Two implementation approaches are comparable only when they receive the same requirement and tests. Keep the attempts in isolated branches, then assess behavior, complexity, and verification evidence side by side. The fastest completion should not outweigh a result that better meets the shared acceptance conditions.

Separate implementation from review

Separate implementation from review with a fixed handoff revision and explicit acceptance conditions. Give the reviewer that code baseline and a focused question, then connect each finding and resolution to the revision being considered. This prevents review comments from drifting onto code that changed after the assessment began.

Work on distinct repository areas

Parallel repository work needs separable ownership boundaries, shared-interface rules, and an integration owner. Allocate worktrees or branches by those boundaries and agree on merge order before changes overlap. Joint checks at integration should verify that individually valid contributions still work together.

Keep a failed parallel attempt bounded

One agent's unsuccessful change should be assessed within its assigned files and revision rather than allowed to disturb unrelated work. Compare the branch result with the task brief, retain useful evidence, and decide separately whether to revise, defer, or discard that attempt. Use Orca's worktree coordination as a reference while keeping engine permissions and final integration decisions explicit.

Prepare Orca engines and remote access

Each agent engine needs its own installation, authentication, and execution policy checked. Review remote or companion access separately from worktree isolation: independent file branches do not provide provider subscriptions or a security boundary for the processes running within them.

  • Confirm each agent’s setup

    Install and authenticate the engines you intend to use. Source availability does not include provider subscriptions or unlimited model usage.

  • Inspect execution and remote access

    Review the chosen engine’s approval settings and any SSH or companion connection. A worktree is file isolation, not a security boundary.

Integrate one reviewed Orca result

Parallel changes need more than individual agent completion before they can be combined. Review the proposed revisions, test the accepted set together, and make a separate merge decision so integration evidence reflects the actual version that would enter the repository.

  1. Integrate one reviewed result

    Choose the accepted change, run checks on the combined revision, and keep the merge decision explicit rather than equating task completion with approval.

Related Software Projects

Compare the workflows, operating requirements, and source licenses of these projects.

Orca FAQ

Which Orca project does this page describe?
The source is stablyai/orca, the high-star multi-agent development workspace. Other repositories named Orca are separate projects and should not be used as interchangeable installation sources.
Does Orca replace Codex or Claude Code?
It organizes existing coding agents in a shared workbench. The selected engines still perform the model-driven work and retain their own account requirements.
Do separate worktrees make parallel work safe automatically?
They isolate working files, but do not provide a complete security boundary or resolve shared-interface conflicts. Define permissions, task scope, and integration checks separately.
Can agents use different models or providers?
The project supports multiple CLI agent engines. Exact models, quotas, and authentication depend on the selected engine and account.
What is the mobile app for?
The documented companion lets a user monitor and steer work associated with a desktop host. It does not mean every provider engine runs directly on the phone.
Should every task be sent to several agents?
No. Use parallel attempts when independent work or comparison is valuable. Shared mutable work and unclear acceptance criteria can add coordination cost without improving the result.